2026-10-06 20:02:19
Welcome to the Premium edition of How They Make Money.
Over 300,000 subscribers turn to us for business and investment insights.
In case you missed it:
OpenAI used DevDay last week to unveil more than 20 products and features. The most important announcement was Dots.
Dots are always-on AI agents with their own cloud computer, access to more than 4,000 apps, and the ability to keep working toward your goals even when you are not actively prompting them. Sam Altman described the shift as a “whole new way to work with AI” and a new set of “superpowers.”
The contrast with Meta Connect was striking. Meta showed Muse booking trips, shopping, and handling everyday tasks. OpenAI showed Dots managing projects, coding, preparing documents, coordinating across Slack and Teams, and helping run a startup.
For now, the two companies are attacking the agent market from opposite ends.
Meta is pushing toward AI for everyone.
OpenAI is starting with AI for high-value work.
And that difference may explain where ChatGPT is heading next.
Today at a glance:
The pieces start to fit together
OpenAI wants to become the operating layer
The business model gets more interesting
The cost of autonomy
Who owns the workday?
Some of the most interesting opportunities emerge in smaller, undercovered businesses that institutions and analysts barely follow.
I’m working on a new way to help you find them before they become obvious.
Spots will be strictly capped at launch.
OpenAI has spent the past year turning ChatGPT into much more than a chatbot, but the pieces have not always fit together cleanly.
ChatGPT started as the conversational interface.
Codex became the place for agentic coding, organized around projects and files.
ChatGPT Work extended that model to broader knowledge work, adding connections to SaaS apps and a persistent cloud environment.
More recently, OpenAI has started pulling those experiences into the same product.
The result is powerful, but increasingly complicated. You can have several projects, agents, threads, apps, and tasks running at once. At some point, the bottleneck becomes the person trying to keep track of it all.
That’s where Dots fit.
A Dot sits above your individual tools as a coordinating layer. It remembers what you are working on, follows projects over time, and can coordinate work across them. It acts like a control tower for your AI work.
The shift matters because the more work OpenAI hosts on its own cloud computers, the more persistent the relationship becomes. A chatbot conversation is relatively easy to move between providers. Recreating a collection of projects, files, permissions, preferences, and ongoing agents somewhere else is much harder.
In other words, Dots could make ChatGPT both easier to use and harder to leave.
2026-10-03 22:00:55
Welcome to the Saturday PRO edition of How They Make Money.
Over 300,000 subscribers turn to us for business and investment insights.
In case you missed it:
📊 Monthly reports: 200+ companies visualized.
📩 Tuesday articles: Exclusive deep dives and insights.
📚 Access to our archive: Hundreds of business breakdowns.
📩 Saturday PRO reports: Timely insights on the latest earnings.
Today at a glance:
🌐 Accenture: AI Creates More Work
👟 Nike: Still Going Downhill
🛳️ Carnival: Europe Comes Back
🎿 Vail Resorts: Pass Sales Slide
Accenture Q4 revenue rose 6% Y/Y to $18.7 billion ($660 million beat), with GAAP EPS of $3.29 ($0.11 beat).
Demand rebounded to a new high. New bookings grew 5% in local currency to $22.2 billion, reversing last quarter's decline. Managed services bookings hit a record $12.8 billion. Shares surged nearly 20% initially, their best post-earnings move on record.
So far, AI is creating more work for Accenture than it is destroying. Consulting revenue grew 6% (or 7% in local currency), and nearly 100 additional clients started their first advanced AI projects during Q4, bringing the FY26 total above 400. Management says much of the demand is coming from companies building the data foundations, digital core, and AI infrastructure needed before deploying agents at scale.
The economics are changing, though. Accenture acknowledged lower pricing in many parts of the business as AI improves productivity and competition intensifies. Hiring will also slow in FY27 as AI changes delivery. So the disruption thesis isn’t disappearing, but it’s showing up first in pricing and labor intensity rather than collapsing demand.
M&A remains another major growth lever. Accenture deployed $1.9 billion in Q4. The previously announced OT-security deals have now closed, and management expects roughly $5 billion of acquisitions in FY27. It also plans to return at least $9.5 billion to shareholders.
For FY27, Accenture guided to 3%–6% revenue growth in local currency. That said, acquisitions are expected to contribute 2%–2.5% to growth, so the organic outlook is considerably more subdued than the headline range.
Takeaway: The quarter doesn’t settle the AI disruption debate, but it brings some much-needed good news. Bookings rebounded, and more enterprises are hiring Accenture to build the infrastructure required for AI. The real pressure may increasingly show up in pricing and headcount rather than demand disappearing altogether.
Nike Q1 revenue fell 4% Y/Y to $11.2 billion ($110 million miss), while EPS slipped 2% to $0.48 ($0.04 beat). Gross margin improved 60 bps to 42.8%, thanks to lower warehousing and logistics costs.
Nike’s performance portfolio remains encouraging. Running, football, training, and basketball collectively grew at a high-single-digit rate, suggesting Elliott Hill’s Sport Offense is gaining traction. But competitive pressure has intensified. Kylian Mbappé recently ended a two-decade relationship with Nike to join On (visualized here), which is now using him to launch its own football business in 2027.
North America was also the only major geography to grow. Elsewhere, the reset remains painful. Nike Sportswear fell by low double digits, while management acknowledged it had oversupplied Jordan retros and is now deliberately reducing launch volume to restore scarcity. NIKE Direct declined 8%, including a 13% drop in digital, while wholesale slipped 1%.
Greater China revenue fell 22% (or 26% in constant currency) to $1.18 billion, with wholesale down 28%. Nike is intentionally reducing promotions and narrowing its digital distribution there, which should improve brand health eventually. But it also means the reset will weigh on revenue for longer.
2026-10-02 20:00:54
Welcome to the Free edition of How They Make Money.
Over 300,000 subscribers turn to us for business and investment insights.
In case you missed it:
Anthropic’s IPO prospectus hasn’t officially landed yet, but Reuters obtained a leaked draft this week, giving us the first real look inside one of AI's most extraordinary growth stories.
A few numbers already stand out:
Revenue reportedly reached ~$4.6 billion in 2025, up 12x Y/Y.
Compute and infrastructure spending topped $7 billion.
Nearly half of revenue came through AWS and GCP, showing how Amazon and Google are simultaneously investors, infrastructure providers, distributors, and competitors.
Much more is hidden beneath those numbers, including customer concentration and cash burn. Anthropic has also agreed to spend at least $518 billion on future compute and infrastructure over the next decade, with roughly 80% of those commitments reportedly non-cancellable. I’m waiting for the public S-1 before doing the full teardown. Stay tuned for Anthropic visualized! 📊
Today at a glance:
🏢 Meta’s enterprise push
☁️ Micron raises the floor
Most of the businesses I cover are already firmly on investors’ radar.
But there’s another part of the market I don’t cover nearly enough.
I’ve been working on something new that fills that gap.
Access will be limited when it opens. Join the priority list to hear it first.
Meta spent the past few weeks showing what AI could mean for consumers.
This week, it turned to businesses.
On Monday, Meta unveiled Meta Enterprise Platform, which Zuck called the company’s “next major pillar.” The new division will bring several pieces of Meta’s AI stack under one roof for businesses and developers, including:
Muse, its personal AI agent
Meta Business Agent
Muse API
Muse Code
Meta’s underlying models and infrastructure
Meta also hired MongoDB CEO CJ Desai to lead the effort. Desai previously ran product and engineering at Cloudflare and spent nearly eight years at ServiceNow. He brings Meta deep enterprise software experience at the top, something it has historically lacked.
The announcement was ambitious enough to rattle the software sector. Salesforce, ServiceNow, Oracle, Adobe and other enterprise names sold off as investors contemplated another hyperscaler moving up the software stack.
That reaction may prove premature.
Building the technology is only part of enterprise software. Large customers also require long sales cycles, procurement, security reviews, integrations and implementation support. Meta has spent decades building consumer products and advertising tools, not selling mission-critical software to CIOs.
That’s why Tuesday’s announcement may be more interesting.
Meta introduced Muse for Small Business, allowing companies to connect the agent to tools they already use, including Shopify, QuickBooks, Stripe, Canva, Slack, Notion, Figma, and Zoom, alongside their Facebook and Instagram business accounts.
The opportunity is not necessarily to replace those applications. It’s to sit above them.
A small business owner could ask Muse to understand why sales slowed, identify customers worth targeting, prepare new creative, and coordinate the work across several apps. Meta says actions involving publishing, sending, or spending still require approval.
That’s a much more natural fit for Meta. The company already works with more than 200 million businesses, primarily because they want access to customers on Facebook and Instagram. Many are small enough that the owner is simultaneously the marketer, operator, salesperson, and finance department (like yours truly).
An AI agent that can help across all four functions could be genuinely useful.
It also extends the same thesis we discussed last week without requiring Meta to win an entirely new market.
For consumers, Muse wants to become the interface to the internet
For businesses, it wants to become the interface to the software stack
The applications underneath may remain the same. What changes is who controls the layer where users express intent.
Meta is also pushing this aggressively for financial reasons.
The company is spending more than $100 billion on AI infrastructure this year. In previous earnings calls, almost every path to monetization eventually came back to advertising, leaving investors skeptical about how broadly Meta could monetize that spending. Enterprise software gives Meta another way to earn a return on the same models, agents, and compute.
Its AI monetization stack is quickly expanding:
Advertising monetizes attention
Subscriptions monetize heavy users
Commerce could monetize transactions
Business software could monetize the AI stack itself
💡 Takeaway: None of this means Meta suddenly becomes Salesforce or ServiceNow. But it is starting to turn its massive AI investment into businesses that extend beyond advertising. Enterprise is the boldest ambition, while small business looks like the lowest-hanging fruit given Meta’s existing customer relationships and distribution.
Micron closed fiscal 2026 with another extraordinary quarter.
Revenue surged 379% Y/Y and 31% Q/Q to $54.2 billion ($2.7 billion beat).
Adjusted EPS: $33.42, up from $3.03 a year ago ($1.60 beat).
Gross margin: 87%, up from 85% in Q3.
Operating cash flow: $44.0 billion, while free cash flow reached $33.2 billion.
Q1 FY27 guidance: Revenue of ~$61.5 billion (+351% Y/Y; ~$3.6 billion beat) and adjusted EPS of ~$38.15 (~$2.23 beat).
The scale is becoming difficult to process. Micron generated $133 billion of revenue in FY26, up from $37 billion a year earlier, while adjusted gross margin jumped to 81%. It ended the year with nearly $74 billion of cash and investments after producing more than $62 billion of adjusted free cash flow.
The growth is broadening beyond HBM. Core Data Center revenue reached $18.0 billion in Q4, up 56% sequentially, while Cloud Memory climbed to $16.3 billion. Even Mobile and Client posted a remarkable 90% gross margin, showing how far the memory shortage has spread beyond AI accelerators.
The key update was management’s view that the memory shortage is worsening.
Micron now expects supply-demand conditions in fiscal 2027 and 2028 to be tighter than they were in 2026, with no clear line of sight to when the market returns to balance. That matters because the biggest risk after this kind of earnings explosion is usually obvious: supply catches up, pricing rolls over, and the memory cycle turns.
Micron is also making more progress on the multi-year contracts we discussed last quarter. It now has 26 Strategic Customer Agreements, up from 16, with roughly $32 billion of customer commitments. These agreements cover more than 35% of revenue through 2030, and about three-quarters of that revenue includes defined pricing frameworks. The goal is to make the next downcycle less violent than the last one.
At the same time, Micron is spending aggressively to add capacity. CapEx reached $27 billion in FY26 and is expected to rise further, including roughly $25 billion in the first half of FY27 alone. But much of that new clean-room capacity will not arrive until 2028 and beyond, which helps explain why management still sees such a tight market.
Despite the extraordinary numbers, Micron still trades at a single-digit forward earnings multiple. The market is effectively saying these margins and earnings cannot last.
That skepticism is understandable. Memory has always been cyclical, and an 87% gross margin looks like the kind of number investors normally associate with a peak.
But Micron is trying to change the evidence investors use to make that judgment:
Long-term customer agreements are replacing annual handshakes.
Customers are putting down billions in deposits and guarantees.
Supply remains constrained well into 2027 and 2028.
AI is increasing memory intensity across data centers, PCs, phones, and eventually physical AI.
The key question is no longer whether Micron is having an extraordinary cycle. It clearly is. The question is whether this cycle has become structurally different enough to deserve a different multiple.
💡 Takeaway: Micron just delivered the kind of quarter that would normally scream peak cycle. Yet supply is getting tighter, customer commitments are getting longer, and FY27 guidance is still moving higher. The market is still pricing Micron like memory eventually reverts to the old playbook. The question is whether AI has changed that playbook enough to justify a different multiple. Just remember the four most dangerous words in investing: “this time is different.”
That's it for today.
Happy investing!
P.S. I’m working on something new for investors looking beyond Big Tech.
Disclosure: I own AMZN, GOOG, and META in App Economy Portfolio. I share my ratings (BUY, SELL, or HOLD) with App Economy Portfolio members.
Author's Note (Bertrand here 👋🏼): The views and opinions expressed in this newsletter are solely my own and should not be considered financial advice or any other organization's views.
2026-10-01 04:47:26
Welcome to the Premium edition of How They Make Money.
Over 300,000 subscribers turn to us for business and investment insights.
🔥 The September report is here!
All the key earnings visuals from the past month in one report.
✔️ Cut through the noise with clear, concise financial snapshots.
✔️ See revenue trends, profit margins, and key takeaways instantly.
We visualized 200+ companies this season:
In case you missed it:
Download the full report below or log in to your account.
Your voice matters! Help us shape future reports. Got a company or sector you're curious about? Hit 'Reply' and let us know!
Here’s a sneak peek. 👀
Inside the September report:
⚙️ Semis: Broadcom, Micron.
🤖 AI infra: Dell, Oracle, Nscale, SB Energy.
🛡️ Cybersecurity: Palo Alto Networks, Zscaler.
📊 Data: C3.ai, HPE, MongoDB, Rubrik, Samsara, Snowflake.
🖥️ Workflow: Adobe, DocuSign, GitLab, UiPath.
👟 Consumer: Costco, Lululemon, Oura.
😎 Travel & leisure: Carnival, Vail Resorts.
And more, including Darden, Didi, Meituan, FedEx, GameStop, and Wealthfront.
2026-09-29 20:03:19
Welcome to the Premium edition of How They Make Money.
Over 300,000 subscribers turn to us for business and investment insights.
In case you missed it:
Nscale was spun out barely two and a half years ago.
It has already signed contracts worth up to $103 billion in future revenue.
Now it wants to build a full-stack AI hyperscaler from power to token.
The company secures power and land, builds data centers, buys the GPUs, sells the compute, and is now pushing further into AI software. And it’s scaling at an almost absurd pace.
Yet today, only 25,000 GPUs are active, compared with 461,000 active and contracted GPUs. Revenue reached just $141 million in the first half of 2026, even after growing 1,252% Y/Y.
The gap between what Nscale operates today and what it has already contracted is the main IPO story.
Can it turn extraordinary demand into a vertically integrated AI infrastructure platform before the capital requirements overwhelm the economics?
And then there is NVIDIA, which somehow shows up on almost every side of the transaction.
I condensed the S-1 into a clean breakdown, supported by our signature visuals. By the end, you’ll have a clear view of the Nscale investment case.
Today at a glance:
Overview
Business model
Financial highlights
Risks & challenges
Management
Use of proceeds
Future outlook
Personal take
Nscale was born out of a bitcoin miner.
Before the company was spun out in May 2024, its team was part of Arkon Energy, which had spent years assembling powered sites in markets where electricity was cheap and abundant.
The opportunity changed when AI demand exploded. The logic was simple. If models kept getting larger and more compute-intensive, power would become the bottleneck.
GPUs get most of the attention, but they are useless without enormous amounts of electricity behind them.
Nscale started by securing land and power in markets including Norway, Iceland, Portugal, and the US, then building compute capacity on top. Management now calls access to large blocks of reliable power the primary constraint on AI infrastructure.
As of August, Nscale had:
5 active and 12 contracted data-center sites
1.37 GW of active and contracted capacity
Line of sight to roughly 10 GW of potential power capacity
The biggest example is Monarch, a 2,250-acre campus in West Virginia acquired in March. Nscale says the site could eventually support more than 8 GW of power, including over 6.5 GW of IT load. The first 1.37 GW is expected to be online in the first half of 2028.
Nscale does not want to stop at power and data centers.
It also owns and operates the GPUs inside them, then sells the compute through its cloud platform.
The planned acquisition of Anyscale, which builds software to run and orchestrate AI workloads, pushes Nscale another step higher into the software layer.
That is how Nscale gets to its pitch of becoming a full-stack AI hyperscaler.
But the current business still looks very different from that end state. Most of Nscale’s active capacity today sits in colocation or leased facilities, which lets the company get GPUs online while its own campuses are being built. By contrast, most of its contracted capacity is expected to sit in facilities Nscale owns.
Nscale is already selling the future version of the company while much of the physical infrastructure behind it still needs to be built.
Takeaway: Nscale’s original advantage was access to power. Now it’s trying to use that foothold to own more of the economics around AI compute, from the data center to the GPU to the software layer. The opportunity is enormous, but the vertically integrated company investors are being asked to value is still mostly under construction.
Nscale monetizes its infrastructure through two businesses:
🖥️ Nscale Infrastructure: Reserved AI compute sold under long-term contracts to hyperscalers, frontier labs, and large technology companies.
☁️ Nscale Cloud: On-demand compute and higher-value platform services aimed at AI-native companies, enterprises, and sovereign customers.
Today, the first business drives almost all of the economics.
2026-09-26 22:03:52
Welcome to the Saturday PRO edition of How They Make Money.
Over 300,000 subscribers turn to us for business and investment insights.
In case you missed it:
📊 Monthly reports: 200+ companies visualized.
📩 Tuesday articles: Exclusive deep dives and insights.
📚 Access to our archive: Hundreds of business breakdowns.
📩 Saturday PRO reports: Timely insights on the latest earnings.
Today at a glance:
🛒 Costco: Refunds Fund Price Cuts
🍪 General Mills: Innovation Gains Traction
🫒 Darden: Olive Garden Still Lags
Costco Q4 revenue rose 11% Y/Y to $95.7 billion ($0.8 billion beat), while EPS increased 15% to $6.75. That included a $0.15 benefit from tariff refunds, so underlying EPS growth was closer to 12%. Comparable sales rose 9.4%, or 6.7% excluding gas and FX, essentially maintaining Q3’s 6.6% pace.
Fuel remained a tailwind, adding roughly 3 points to comps, but the core business was balanced. Adjusted traffic increased 3.3%, and average ticket rose 3.3%. Digitally enabled comps grew another 20%, with annual digitally enabled sales now exceeding $33 billion (~11% of Costco’s overall FY26 revenue).
Costco received $184 million in tariff refunds during Q4 and reinvested part of that into lower prices across produce, meat, beverages, furniture, and other categories. Management already received a similar amount in Q1 and plans to return most future refunds to members through better value.
Membership remains healthy, but growth continues to normalize. Paid members increased 4% to 84.1 million, while Executive memberships grew 9% to 42.3 million. US/Canada renewal improved to 92.3%. Costco also plans 33 warehouse openings in FY27, accelerating from 25 net additions in FY26.
Consumer insights from Numerator data show that Costco’s recent household growth is skewing toward Gen Z and lower-income shoppers, while lower-income customers are already making fewer trips and spending less per basket.
The stock is still expensive. Costco’s forward P/E has fallen from above 50x at its peak to roughly 41x, but that still leaves little room for error.
Bottom Line: The gasoline boost still hasn’t normalized, but Costco’s core comps are holding around 7% anyway. With membership growth returning to a more normal pace, the next leg of growth increasingly depends on getting more from each member through digital, pharmacy, Executive upgrades, and a broader services ecosystem. The market is certainly counting on it.